LAUSR

Abstract An experimental firefighter-wearable device exploiting an Artificial Intelligence (AI) based algorithm is described, with the purpose of pre-warning wearers of dangerous rates of temperature rise in the environment, for example ahead of flashover. This exploits a compact embedded Artificial Neural Network (ANN) with a second stage classifier, reading temperature data from the in-built thermocouple, to produce a 30 s output window of predicted temperatures, used to inform pre/full alarm states and a series of audible/visual warnings. The algorithm and device have been tested in two controlled fire behaviour training environments in both the UK and US during multiple flashovers, demonstrating the integrity of the predictions and subsequent alarm triggering in this environment, specifically, triggering an alarm 27 s ahead of flashover for a unit positioned at firefighter crouching height (0.8 m). These experiments included testing on a reference mannequin to explore the potential impact of the boundary effect on temperature measurements close to a tunic, as well as the impact of simulated firefighter movement on the output predictions and therefore alarm states. The results from subsequent testing of the devices in realistic firefighting conditions by firefighter instructors demonstrate the dynamic response of the algorithm to changing conditions equivalent to a serious domestic building fire, including the potential benefit of the precautionary ‘pre-alarm’. Finally, a snapshot of results from extended trials in an uncontrolled training environment with multiple firefighters is discussed, including early developments towards a potential concept of use including cloud-based aggregation of individual firefighter temperature exposure data.